Menu

← Leaderboard

gpt-oss-20b with Hermes Agent

MXFP4 (official GGUF) on llama.cpp commit a37cc46. Tested

Task completion
90% (27/30)
Tool-call success
77.5%
Decode tok/s
79.4
Prefill tok/s
4999.6
Server memory
12.5 GB
Avg turns
5.3

Setup

Model
gpt-oss-20b
Family / maker
gpt-oss · OpenAI
Parameters
20B
Architecture
MoE
Weights
ggml-org/gpt-oss-20b-GGUF
Quant
MXFP4 (official GGUF)
Engine
llama.cpp commit a37cc46
Hermes Agent
v0.22.4
Tested
2026-10-22
Context
64K (65,536 tokens); long-context task t09 at 128K or 192K
Memory tier
≤16 GB (measured)

Per-task results

  • t01: 3/3
  • t02: 3/3
  • t03: 3/3
  • t04: 3/3
  • t05: 1/3
  • t06: 3/3
  • t07: 3/3
  • t08: 2/3
  • t09: 3/3 (rerun 2026-10-23, 128K context)
  • t10: 3/3

Copy-paste config

1. Start the model server (llama.cpp)

Exact command used in our test. Download the GGUF from ggml-org/gpt-oss-20b-GGUF into ./models/.

llama-server --model ./models/gpt-oss-20b-MXFP4.gguf --alias gpt-oss-20b --host 127.0.0.1 --port 18080 --ctx-size 65536 --parallel 1 --n-gpu-layers 999 --flash-attn on --jinja

2. Point Hermes Agent at it

hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:18080/v1
hermes config set model.default gpt-oss-20b

Other gpt-oss setups